Evolution
Evolution answers two questions about any artifact: where did this come from, and what would I break if I changed it.
Lineage and blast radius
WHERE IT CAME FROM WHAT IT INFLUENCES
(dependencies) (dependents)
Provider ──┐ ┌── Evaluation A
Dataset ──┼──► THIS ARTIFACT ──────┼── Evaluation B
Prompt ──┘ └── Benchmark C
│
transitive impact:
everything downstream
of thoseBefore you change something, you should know what it will affect. Evolution makes that explicit rather than leaving it to memory — and memory is exactly what fails when a dataset is regenerated three months after anyone touched it.
What the tab holds
Verdict banner — impact before inventory. The first thing you see is what a change here would mean, not a list of rows.
Where it came from — the artifact's dependencies.
What it influences — its dependents, and the transitive impact count.
Deployment timeline (AI Git) — every revision, which is promoted, rollback readiness, and the recorded rationale for each change. See AI Git.
How it changed — the historical revisions.
Evidence — the evaluations that bear on this artifact.
Where the relationships come from
They are real, not declared. The graph is built from actual references: an evaluation that pins a prompt genuinely depends on it; an agent configured with a provider genuinely reaches it. Nobody maintains a dependency file, so nothing can be out of date.
Impact as a number that means something
The impact count appears throughout the platform — in Brok answers, in investigations, in Evolution itself — and it is always accompanied by a sentence saying what it means:
"Promoting it changes what 3 downstream artifacts depend on." "Nothing downstream depends on this artifact, so the promotion is contained."
A bare number invites misreading. A number with its consequence stated is engineering information.
Questions Evolution answers
- Can I safely regenerate this dataset? — check the dependents.
- Why does this evaluation exist? — check its dependencies.
- Is this prompt still used anywhere? — an artifact with no dependents may be dead weight.
- What is the blast radius of this change? — the transitive impact.
- Is production running the newest revision? — the deployment timeline.
Brok reaches the same reading with "Show every artifact affected by X" and "What changed between these revisions?".
See also: AI Git · Forge Graph · Registry